Abstract (eng)
I investigated picture-object recognition in kea (Nestor notabilis) in three different experiments by asking subjects to transfer a learned behavior (a discrimination task) from pictures to objects, and vice versa.
Experiment one investigated whether kea could transfer a two-choice discrimination task between pictures displayed on a touchscreen computer and real objects presented on the aviary ground. Results showed strong evidence of object-to-picture transfer in the first few trials of the object-to-touchscreen test session. However, there was no evidence from this experiment for picture-to-object transfer. This was because kea performed significantly better on the object discrimination task than on the touchscreen, so it was impossible to tell whether transfer had occurred, or whether the subjects had simply learned the object task “from scratch”. The conclusions for experiment one are twofold. First, kea showed picture-object recognition in the form of transfer of a discrimination task from objects to the touchscreen. Second, as kea very quickly reached significant positive results in a real-object discrimination task, direct discrimination transfer from pictures to objects was not a demanding enough task to test for recognition.
Experiment two addressed outstanding concerns from experiment one. To tackle the lack of evidence for picture-to-object transfer, a reversal-learning procedure was implemented, which is known to be more difficult for kea to solve. To address the concern that object-to-picture transfer only occurred in the first few trials of the touchscreen test session, an operant conditioning box which mimicked the touchscreen was used to test the object condition. This box would hopefully reduce distracting factors, and make it possible to attain a transfer detectable throughout the entire test session. In fact, results suggest that these procedural modifications were successful, and showed a strong transfer from picture to object which was detectable over the entire test session. However, the operant conditioning box itself proved to be somewhat of a hindrance, as only three subjects of the original thirteen that began training learned to use the box correctly.
Experiment three tested whether kea could transfer between objects and framed photographs, a medium with which they had never worked with before. As in experiment two, procedural variations between mediums were minimized to increase the probability of showing robust transfer. Furthermore, both photographs and objects were presented to subjects in a way they were already familiar with, so a sample size was gained which more accurately represented the group as a whole. Results showed picture-to-object recognition in the form of negative transfer in the first session of the photo-to-object condition. This supported similar results found in experiment two, but with a larger sample size. When taken together with the results from experiment one, it is clear that kea can transfer a learned discrimination task both from pictures to objects, and from objects to pictures.
Based on the results described in this thesis, it is now known that kea can correspond objects and pictures, both when these are presented on a touchscreen and as photographs. Furthermore, the fact that kea were successful in recognizing printed photographs, despite being naïve to these, implies that they are quite flexible in their abilities to recognize different types of pictures. This lends support to the use of pictures in cognitive research with this species.